Irregular pulse classification method of 12-lead electrocardiogram signal and device for performing the same

Abstract

To provide an irregular pulse classification method of a 12-lead electrocardiogram signal and a device for performing the method.SOLUTION: A device for performing irregular pulse classification of a 12-lead electrocardiogram signal according to one embodiment of the present invention is a computing device provided with one or more processors and a memory for storing one or more programs executed by the one or more processors. The device includes a data input module which receives input of the 12-lead electrocardiogram signal and a classification module which outputs a classification result about the 12-lead electrocardiogram signal by using the technique of a machine learning base based on the input 12-lead electrocardiogram signal.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] SUMMARY OF THE INVENTION Embodiments of the present invention relate to techniques for classifying arrhythmias from electrocardiogram signals. [Background technology]

[0002] According to the World Health Organization (WHO), heart disease kills millions of people worldwide each year. Electrocardiography, a well-known method for testing for heart disease, is a non-invasive test that uses electrodes attached to the skin to sense and record the electrical activity of the heart during the cardiac cycle. Typically, 12 ECG signals are acquired and analyzed using 10 electrodes attached to the arms, legs, and chest to determine the presence or absence of cardiac disorders such as arrhythmia.

[0003] Traditionally, doctors have diagnosed cardiac diseases through recordings of electrocardiogram signals for each patient, but recently, there have been increasing attempts to use deep learning to analyze and diagnose electrocardiogram signals.

[0004] Conventional deep learning techniques for ECG signal analysis and diagnosis classify cardiac arrhythmias by extracting time-frequency features from single-lead ECG signals. However, some cardiac arrhythmias are only observed in specific ECG channels, which may reduce classification accuracy depending on the type of cardiac arrhythmia being classified.

[0005] Therefore, accurate diagnosis of cardiac arrhythmia requires comprehensive and thorough confirmation of ECG signals obtained from 12 channels. That is, since ECG signals measured through 12 leads vary depending on the type of arrhythmia, comprehensive analysis of ECG signals obtained from 12 channels is necessary to accurately determine a patient's arrhythmia. [Prior art documents] [Patent documents]

[0006] Korean Patent Registration No. 10-2163217 (2020.09.29.) Summary of the Invention [Problem to be solved by the invention]

[0007] Embodiments of the present invention are directed to providing a method for classifying arrhythmias from standard 12-lead electrocardiogram signals using machine learning techniques. [Means for solving the problem]

[0008] According to an exemplary embodiment of the present invention, there is provided a computing device having one or more processors and a memory storing one or more programs executed by the one or more processors, the computing device including: a data input module that receives an input of a 12-lead electrocardiogram signal; and a classification module that outputs a classification result for the 12-lead electrocardiogram signal using machine learning-based techniques based on the input 12-lead electrocardiogram signal.

[0009] The classification module may include an artificial neural network model that receives the 12-lead electrocardiogram signal and is trained to classify arrhythmia information based on the 12-lead electrocardiogram signal.

[0010] The artificial neural network model may include an initial feature block that outputs an initial feature map through a convolution operation based on the input 12-lead ECG signal; an attention block that outputs a concentrated feature map through an element-by-element weight operation based on the initial feature map output from the initial feature block; a residual block that outputs a deep feature map through a shortcut operation based on the initial feature map output from the initial feature block; a combination block that combines the concentrated feature map output from the attention block and the deep feature map output from the residual block to output a final feature map; and a classification block that classifies the type of arrhythmia based on the final feature map output from the combination block.

[0011] The attention block may perform max pooling and average pooling in parallel on the initial feature maps output from the initial feature block through a max pooling layer and an average pooling layer, combine the pooling results output from the max pooling layer and the average pooling layer through a first combined layer to output a weight feature map, and combine the initial feature map and the weight feature map through a second combined layer to output a concentrated feature map.

[0012] The residual block is configured to include N (N is a natural number equal to or greater than 1) short residual blocks sequentially connected to reflect the features of the initial feature map output from the initial feature block. The N short residual blocks receive a previous feature map (a feature map output from the (N-1)th short residual block), output a new feature map from the previous feature map through a convolutional layer, and combine and output the previous feature map and the new feature map through a third combined layer.

[0013] The first and third combined layers may use element-wise sums, and the second combined layer may use element-wise multiplications.

[0014] The classification block may perform global max pooling and global average pooling in parallel on the final feature map through a global max pooling layer and a global average pooling layer, and may concatenate the pooling results output from the global max pooling layer and the global average pooling layer through a concatenation layer.

[0015] According to another exemplary embodiment of the present invention, there is provided a method for arrhythmia classification of a 12-lead electrocardiogram signal, the method being performed on a computing device having one or more processors and a memory storing one or more programs executed by the one or more processors, the method including: receiving an input of a 12-lead electrocardiogram signal; and outputting a classification result for the 12-lead electrocardiogram signal using a machine learning-based technique based on the input 12-lead electrocardiogram signal. [Effects of the Invention]

[0016] According to an embodiment of the present invention, arrhythmias are classified from standard 12-lead electrocardiogram signals using machine learning techniques, which advantageously allows for rapid and accurate classification from electrocardiogram signals. [Brief explanation of the drawings]

[0017] [Figure 1] 1 is a diagram showing the configuration of an arrhythmia classification device for 12-lead electrocardiogram signals according to an embodiment of the present invention.

[0018] [Figure 2] 1 is a diagram illustrating the structure of a classification module constituting an arrhythmia classification device for 12-lead electrocardiogram signals according to an embodiment of the present invention;

[0019] [Figure 3] FIG. 2 illustrates a 12-lead electrocardiogram signal according to one embodiment of the present invention.

[0020] [Figure 4] FIG. 3 is a diagram illustrating an attention block 320 of an arrhythmia classification device for a 12-lead electrocardiogram signal according to an embodiment of the present invention.

[0021] [Figure 5] FIG. 3 is a diagram illustrating a residual block 330 of an arrhythmia classification apparatus for a 12-lead electrocardiogram signal according to an embodiment of the present invention.

[0022] [Figure 6] FIG. 2 illustrates the layer structure of an artificial neural network model in a classification module of an arrhythmia classification apparatus for 12-lead ECG signals according to one embodiment of the present invention.

[0023] [Figure 7] 1 is a flowchart illustrating a method for arrhythmia classification of a 12-lead electrocardiogram signal according to an embodiment of the present invention.

[0024] [Figure 8] FIG. 1 is a block diagram illustrating and describing a computing environment including a computing device suitable for use in exemplary embodiments. DETAILED DESCRIPTION OF THE INVENTION

[0025] Specific embodiments of the present invention will now be described with reference to the drawings. The following detailed description is provided to facilitate a comprehensive understanding of the methods, devices, and / or systems described herein. However, this is for illustrative purposes only, and the present invention is not limited thereto.

[0026] In describing embodiments of the present invention, if a detailed description of known technologies related to the present invention is deemed to unnecessarily obscure the gist of the present invention, such detailed description will be omitted. Furthermore, the terms used below are defined in consideration of their functions in the present invention, and may vary depending on the intentions or practices of users and operators. Therefore, their definitions should be based on the overall content of this specification. Terms used in the detailed description are intended solely to describe embodiments of the present invention and should not be limiting in any way. Unless clearly used otherwise, singular expressions include the plural meaning. In this description, expressions such as "comprises" or "comprises" are intended to indicate certain characteristics, numbers, steps, operations, elements, portions thereof, or combinations thereof, and should not be interpreted to exclude the presence or possibility of one or more other characteristics, numbers, steps, operations, elements, portions thereof, or combinations thereof other than those described.

[0027] In the following description, terms of similar meaning, such as "transfer," "communicate," "send," and "receive" of a signal or information, include not only direct transmission of the signal or information from one component to another, but also transmission via other components. In particular, "transferring" or "sending" a signal or information to one component indicates the final destination of the signal or information, not its direct destination. The same applies to "receiving" a signal or information. In addition, in this specification, "associating" two or more pieces of data or information means that obtaining one piece of data (or information) allows at least a portion of the other pieces of data (or information) to be obtained based on that data.

[0028] Meanwhile, embodiments of the present invention may include a program for performing the methods described herein on a computer, and a computer-readable storage medium containing the program. The computer-readable storage medium may include program instructions, local data files, local data structures, and the like, alone or in combination. The medium may be specially designed and constructed for the present invention, or may be one commonly used in the field of computer software. Examples of computer-readable storage media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and hardware devices specially configured to store and execute program instructions, such as ROMs, RAMs, and flash memories. Examples of the program may include not only machine language code, such as that produced by a compiler, but also high-level language code that can be executed by a computer using an interpreter, for example.

[0029] FIG. 1 is a diagram showing the configuration of an arrhythmia classification device for 12-lead ECG signals according to one embodiment of the present invention, and FIG. 2 is a diagram showing the schematic structure of a classification module constituting the arrhythmia classification device for 12-lead ECG signals according to one embodiment of the present invention.

[0030] 1 and 2, the arrhythmia classification device 100 may include a data input module 200 and a classification module 300. The arrhythmia classification device 100 can classify the type of arrhythmia from a 12-lead electrocardiogram signal using machine learning techniques. Meanwhile, the neural network constituting the arrhythmia classification device 100 according to an embodiment of the present invention may use a 1D convolutional neural network to process the 12-lead electrocardiogram signal.

[0031] In one embodiment, the data input module 200 and the classification module 300 may be implemented using one or more physically distinct devices, one or more processors, or a combination of one or more processors and software, and may not be clearly distinct in their specific operations, unlike the illustrated example.

[0032] The data input module 200 can receive an input of a 12-lead ECG signal. For example, the data input module 200 can acquire a 12-lead ECG signal measured by medical staff. Alternatively, the data input module 200 can acquire a 12-lead ECG signal from an external medical institution or research institute. The 12-lead ECG signal can be used as training data for the arrhythmia classification device 100. When the 12-lead ECG signal is used as training data, arrhythmia information can be labeled. The arrhythmia information can be the type of arrhythmia diagnosed by the 12-lead ECG signal.

[0033] The data input module 200 may provide the input 12-lead electrocardiogram signal to the classification module 300 .

[0034] Meanwhile, as shown in FIG. 3 , a 12-lead ECG signal can be acquired using ten skin surface sensors, including four limb leads (right arm (RA), left arm (LA), right leg (RL), and left leg (LL)) and six chest leads (V1, V2, V3, V4, V5, and V6). The 12-lead ECG signal has a characteristic in that the signal form in each lead varies depending on the type of arrhythmia, and abnormal signs are observed only in specific leads among the 12 leads. Thus, the arrhythmia classification device 100 for a 12-lead ECG signal according to one embodiment of the present invention can classify the arrhythmia type for an input 12-lead ECG signal by learning the correlation between the 12-lead ECG signal and the arrhythmia type using machine learning technology.

[0035] The classification module 300 can classify arrhythmia information related to the 12-lead ECG signal using machine learning-based techniques based on the input 12-lead ECG signal. For example, the classification module 300 can include an artificial neural network model that receives the 12-lead ECG signal from the data input module 200 and is trained to classify arrhythmia information based on the input 12-lead ECG signal. When the 12-lead ECG signal is input, the artificial neural network model can be trained to classify and output arrhythmia information for the 12-lead ECG signal.

[0036] In an exemplary embodiment, the classification module 300 may include an initial feature block 310, an attention block 320, a residual block 330, a sum block 340, and a classification block 350.

[0037] The initial feature block 310 can output an initial feature map through a convolution operation based on the input 12-lead electrocardiogram signal.

[0038] In an exemplary embodiment, the initial feature block 310 may include a convolution layer, an activation layer (PReLU), and an average pooling layer. That is, the initial feature block 310 can output an initial feature map from a 12-lead ECG signal through the convolution layer, the activation layer, and the average pooling layer. Here, the activation layer can use a PReLU (Parametric ReLU) function as an activation function. The activation function is required to readjust the signal strength of a neuron. The PReLU function is a function that multiplies values ​​less than 0 by a parameter (a) adjusted through learning and outputs the result, and outputs values ​​greater than 0 as is.

[0039] The attention block 320 can output a concentrated feature map through element-by-element weight calculation based on the initial feature map output from the initial feature block 310.

[0040] In an exemplary embodiment, as shown in FIG. 4, the attention block 320 may include a convolution layer, a max pooling layer, an average pooling layer, multiple dilated convolution layers, multiple activation layers, and multiple connection layers.

[0041] Specifically, the attention block 320 may simultaneously perform max pooling and average pooling in parallel through a max pooling layer and an average pooling layer. The attention block 320 may also perform an extended convolution operation on each pooling result calculated through the max pooling layer and the average pooling layer, and perform an operation (element-wise sum) to combine the results of the operation through a first connected layer. Here, the value output as the operation result may be a weight feature map (i.e., weight values ​​for each element). The attention block 320 may also perform an operation (element-wise multiplication) to combine the initial feature map and the weight feature map through a second connected layer. The attention block 320 may also output a concentrated feature map through an activation function (ReLU). Here, the attention block 320 may prevent overfitting by using an activation layer (sigmoid function) and dropout to make the weight feature map have weight values ​​for each element ranging from 0 to 1 before operating through the second connected layer.

[0042] That is, the attention block 320 considers the relationship between elements using max pooling, average pooling, and dilated convolution operations, assigns weights to each element depending on its importance, and outputs a weight feature map. The weight feature map and the initial feature map are combined to highlight important elements. Here, the elements may be pixels in the feature map. Meanwhile, in the present invention, the dilation rate of the dilated convolution layer is set to 2, but is not limited thereto.

[0043] The residual block 330 can output a deep feature map through a shortcut operation based on the initial feature map output from the initial feature block 310.

[0044] In an exemplary embodiment, the residual block 330 may include multiple short residual blocks. As shown in FIG. 5, the short residual block may include multiple convolution layers, multiple batch normalization layers, an activation layer (ReLU), and a combination layer.

[0045] That is, the short residual block outputs a new feature map from the previous feature map (the feature map output from the (N-1)th short attention block; in this case, the first short attention block's initial feature map is the previous feature map) through a convolutional layer, and combines (element-wise sum) the previous feature map and the new feature map through a combination layer. Therefore, by combining the new feature map with the previous feature map, each short residual block can smoothly flow information among all short residual blocks in the residual block 330, extracting deep features while solving the problem of feature information disappearance. Meanwhile, overfitting can be prevented by applying dropout to the feature map output from the short residual block activation layer.

[0046] The sum block 340 can output a final feature map by combining the concentrated feature map output from the attention block 320 and the deep feature map output from the residual block 330. In this case, the sum block 340 can use element-wise sum to combine the concentrated feature map and the final feature map.

[0047] The classification block 350 can classify the type of arrhythmia based on the final feature map output from the combination block 340. In this case, the classification block 350 can estimate a probability for each class (arrhythmia information) based on the final feature map, and classify the class with the highest probability as the type of arrhythmia.

[0048] In an exemplary embodiment, the classification block 350 may simultaneously perform global max pooling and global average pooling in parallel through a global max pooling layer and a global average pooling layer. The classification block 350 may also perform an operation of concatenating the pooling results calculated through the global max pooling layer and the global average layer through a concatenation layer. This allows the output data of the separated layers to be combined. Using the global max pooling layer and the global average pooling layer, the final feature map may be output as a preset number of one-dimensional matrix feature maps. The classification block 350 may also estimate probabilities for each class through a fully connected layer using the preset number of one-dimensional matrix feature maps output, and classify the class with the highest probability as the arrhythmia type. A softmax function, for example, may be used as the fully connected layer. The preset number may be the number of arrhythmia types.

[0049] FIG. 6 is a diagram illustrating the layer structure of an artificial neural network model in a classification module of an arrhythmia classification apparatus for 12-lead ECG signals according to an embodiment of the present invention.

[0050] Figure 6 shows the layer structure of the optimized artificial neural network model. Taking the above steps into consideration, we designed a CNN model that can classify 45 arrhythmia types. The main structure of the CNN model is an attention block 320 for highlighting important features and a residual block 330 for improving classification accuracy.

[0051] 7 is a flowchart illustrating a method for classifying arrhythmias in a 12-lead ECG signal according to one embodiment of the present invention. The method illustrated in FIG. 7 can be performed, for example, by the aforementioned arrhythmia classification device for a 12-lead ECG signal. Although the method is described as being divided into multiple steps in the illustrated flowchart, at least some of the steps may be performed in a different order, combined with other steps, omitted, divided into more detailed steps, or with one or more additional steps not illustrated.

[0052] The arrhythmia classification device 100 for a 12-lead ECG signal may receive an input of the 12-lead ECG signal (S710). For example, the arrhythmia classification device 100 for a 12-lead ECG signal may acquire the 12-lead ECG signal measured by medical staff. Alternatively, the arrhythmia classification device 100 for a 12-lead ECG signal may acquire the 12-lead ECG signal from an external medical institution or research institute.

[0053] The 12-lead ECG signal arrhythmia classification device 100 may then output a classification result for the 12-lead ECG signal using machine learning-based techniques S720. For example, the 12-lead ECG signal arrhythmia classification device 100 may include an artificial neural network model that is trained to classify arrhythmia information based on the input 12-lead ECG signal. When the 12-lead ECG signal is input, the artificial neural network model can be trained to classify and output arrhythmia information for the 12-lead ECG signal.

[0054] Finally, the arrhythmia classification device 100 for 12-lead ECG signals may generate arrhythmia information based on the classification result and provide it to a user (eg, medical staff) S730.

[0055] 8 is a block diagram illustrating and describing a computing environment including a computing device suitable for use in the exemplary embodiments. In the illustrated embodiment, each component may have different functions and capabilities in addition to those described below and may include additional components in addition to those described below.

[0056] The illustrated computing environment 10 includes a computing device 12. In one embodiment, the computing device 12 may be an arrhythmia classifier 100 for a 12-lead electrocardiogram signal.

[0057] Computing device 12 includes at least one processor 14, a computer-readable storage medium 16, and a communication bus 18. Processor 14 can cause computing device 12 to operate in accordance with the exemplary embodiments described above. For example, processor 14 can execute one or more programs stored on computer-readable storage medium 16. The one or more programs can include one or more computer-executable instructions that, when executed by processor 14, can be configured to cause computing device 12 to perform operations in accordance with the exemplary embodiments.

[0058] The computer-readable storage medium 16 is configured to store computer-executable instructions or program code, program data, and / or information in other suitable formats. The program 20 stored on the computer-readable storage medium 16 comprises a set of instructions that are executable by the processor 14. In one embodiment, the computer-readable storage medium 16 may be memory (volatile memory, such as random access memory, non-volatile memory, or a suitable combination thereof), one or more magnetic disk storage devices, optical disk storage devices, flash memory devices, or other forms of storage media that can be accessed by the computing device 12 and store desired information, or a suitable combination thereof.

[0059] A communication bus 18 interconnects the processor 14 , the computer-readable storage medium 16 , and various other components of the computing device 12 .

[0060] Computing device 12 may also include one or more input / output interfaces 22 and one or more network communication interfaces 26 that provide interfaces for one or more input / output devices 24. The input / output interfaces 22 and the network communication interfaces 26 are coupled to communication bus 18. The input / output devices 24 may be coupled to other components of computing device 12 via the input / output interfaces 22. Exemplary input / output devices 24 may include input devices such as a pointing device (such as a mouse or trackpad), a keyboard, a touch input device (such as a touchpad or touchscreen), a voice or sound input device, various types of sensor devices and / or photographic devices, and / or output devices such as a display device, a printer, speakers, and / or a network card. The exemplary input / output devices 24 may be included within computing device 12 as a component constituting computing device 12, or may be coupled to computing device 12 as a separate device distinct from computing device 12.

[0061] While the representative embodiments of the present invention have been described above, those skilled in the art will understand that various modifications can be made to the above-described embodiments without departing from the scope of the present invention. Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined by the following claims as well as equivalents thereof. [Explanation of symbols]

[0062] 10: Computing Environment 12: Computing equipment 14: Processor 16: Computer-readable storage medium 18: Communication bus 20: Program 22: Input / output interface 24: Input / output device 26: Network communication interface Arrhythmia Classification Device for 100:12 Lead ECG Signals 200: Data entry module 300: Classification module 310: Initial feature block 320: Caution Block 330: Residual block 340: Bonding block 350: Classification Block

Claims

1. one or more processors; 1. A computing device comprising: a memory for storing one or more programs to be executed by the one or more processors; a data input module for receiving a 12-lead electrocardiogram signal; a classification module that outputs a classification result for the 12-lead electrocardiogram signal using a machine learning-based technique based on the input 12-lead electrocardiogram signal.

2. The classification module:

10. The computing device of claim 1, further comprising an artificial neural network model that receives the 12-lead electrocardiogram signal and is trained to classify arrhythmia information based on the 12-lead electrocardiogram signal.

3. The artificial neural network model comprises: an initial feature block that outputs an initial feature map through a convolution operation based on the input 12-lead ECG signal; an attention block that outputs a concentrated feature map through element-by-element weight calculation based on the initial feature map output from the initial feature block; a residual block that outputs a deep feature map through shortcut operations based on the initial feature map output from the initial feature block; a combination block that combines the focused feature map output from the attention block and the deep feature map output from the residual block to output a final feature map; a classification block that classifies the type of arrhythmia based on the final feature map output from the combination block.

4. The attention block is 4. The computing device of claim 3, wherein the initial feature maps output from the initial feature blocks are subjected to max pooling and average pooling in parallel through a max pooling layer and an average pooling layer, the pooling results output from the max pooling layer and the average pooling layer are combined through a first combined layer to output a weight feature map, and the initial feature maps and the weight feature maps are combined through a second combined layer to output a lumped feature map.

5. The residual block is The initial feature map is configured to include N (N is a natural number equal to or greater than 2) short residual blocks sequentially connected to reflect the features of the initial feature map output from the initial feature block; The N short residual blocks are 5. The computing device of claim 4, wherein a previous feature map (the feature map output from the N-1th short residual block) is input, a new feature map is output from the previous feature map through a convolutional layer, and the previous feature map and the new feature map are combined and output through a third combined layer.

6. the first and third combined layers use element-wise sums; 6. The computing device of claim 5, wherein the second combination layer uses element-wise multiplication.

7. The classification block comprises:

4. The computing device of claim 3, wherein global max pooling and global average pooling are performed in parallel on the final feature map through a global max pooling layer and a global average pooling layer, and the pooling results output from the global max pooling layer and the global average pooling layer are concatenated through a concatenation layer.

8. one or more processors; 1. A method performed on a computing device comprising a memory storing one or more programs to be executed by the one or more processors, the method comprising: receiving an input of a 12-lead electrocardiogram signal; and outputting a classification result for the 12-lead ECG signal using a machine learning-based technique based on the input 12-lead ECG signal.

9. The step of outputting the classification result includes:

9. The method of claim 8, further comprising receiving the 12-lead ECG signals through an artificial neural network model, and classifying arrhythmia information based on the 12-lead ECG signals.

10. The classifying step includes: outputting an initial feature map through a convolution operation based on the input 12-lead ECG signal by an initial feature block; outputting a concentrated feature map through element-by-element weight calculation based on the output initial feature map by an attention block; outputting a deep feature map through shortcut operations by a residual block based on the output initial feature map; Combining the output concentrated feature map and the output deep feature map by a combination block to output a final feature map; 10. The method for classifying arrhythmia of a 12-lead electrocardiogram signal according to claim 9, further comprising the step of: classifying the type of arrhythmia based on the output final feature map by a classification block.

11. The step of outputting the lumped feature map includes: performing max pooling and average pooling in parallel on the initial feature map through a max pooling layer and an average pooling layer; combining the pooling results output from the max pooling layer and the average pooling layer through a first combined layer to output a weight feature map; 11. The method of claim 10, further comprising: performing a combination between the initial feature map and the weight feature map through a second combination layer to output the lumped feature map.

12. The residual block is The initial feature map is configured to include N (N is a natural number equal to or greater than 2) short residual blocks sequentially connected to reflect the features of the initial feature map; The N short residual blocks are 12. The method for arrhythmia classification of a 12-lead electrocardiogram signal according to claim 11, wherein a previous feature map (the feature map output from the (N-1)th short residual block) is input, a new feature map is output from the previous feature map through a convolutional layer, and the previous feature map and the new feature map are combined and output through a third combined layer.

13. the first and third combined layers use element-wise sums; 13. The method of claim 12, wherein the second combined layer uses element-wise multiplication.

14. The step of classifying the type of arrhythmia includes: performing global max pooling and global average pooling in parallel on the final feature map through a global max pooling layer and a global average pooling layer; The method of claim 10 , further comprising: concatenating the pooling results output from the global maximum pooling layer and the global average pooling layer through a concatenation layer.

Images